From: Paul Walker <pwalker@aei-potsdam.mpg.de>
Date: Sat, 22 Nov 1997 19:49:22 +0100 (MET)
To: proj_elliptic@wugrav.wustl.edu
Subject: elliptic/petsc kickoff
Reply: to this message


Hey kids.

So ed asked me to write about the petsc solver. 

Short story: petsc solver is slow, because of one loop doing a
matrix-vector multiply which the compiler does not optimize very well
(theory 1) or that thrashes cache since it is, basically, a daxpy (theory
2).

petsc solver has the following features:

- solves linear equation nabla phi - M phi = N
- has octant support
- has first order excision, so can be used to solve AHBC problems (without
octant currently) where the value of a function is known on the lapse.

Except for the last feature, it is clearly surpassed by BAM elliptic,
which now has octancts.

Questions or comments are welcome.

	- Paul

________________________________________________________________________

  Paul Walker                              <pwalker@aei-potsdam.mpg.de>
  Max-Planck-Institut fuer GravitationsPhysik   Albert-Einstein-Insitut
  Schlaatzweg 1                                  +49-331-27537-55 (tel) 
  14473 Potsdam, GERMANY                         +49 331 27537-98 (fax)

           Question: Have computers improved your life yet?
________________________________________________________________________


elliptic/petsc kickoff / Paul Walker

Created for the WUGRAV CoCoBoard Page Projects Page.
Created by The CoCoBoard.